Improve InstructionCounter and InstructionStatistics

[TheThirdOne]

InstructionCounter could be updated to also have counts for B and J. This would require copying some extra UI elements, adding some state to track counts and changing this if. This would only require understanding the code in InstructionCounter.

rars/rars/tools/InstructionCounter.java

Lines 225 to 237 in 9ff2b88

	// TODO: update this to have labels for the extra formats
	try {
	ProgramStatement stmt = Memory.getInstance().getStatement(a);
	BasicInstruction instr = (BasicInstruction) stmt.getInstruction();
	BasicInstructionFormat format = instr.getInstructionFormat();
	if (format == BasicInstructionFormat.R_FORMAT)
	counterR++;
	else if (format == BasicInstructionFormat.I_FORMAT)
	counterI++;
	else if (format == BasicInstructionFormat.S_FORMAT || format == BasicInstructionFormat.B_FORMAT)
	counterS++;
	else if (format == BasicInstructionFormat.U_FORMAT || format == BasicInstructionFormat.J_FORMAT)
	counterU++;

InstructionStatistics is was never updated from MIPS. Updating this requires more work that InstructionCounter. You will need to update getInstructionCategory(...); that will probably involve a fair amount of instanceof checks. This requires understanding code in rars.riscv.instructions.

rars/rars/tools/InstructionStatistics.java

Lines 238 to 274 in 4b2d4b6

	protected int getInstructionCategory(ProgramStatement stmt) {
	int opCode = stmt.getBinaryStatement() >>> (32 - 6);
	int funct = stmt.getBinaryStatement() & 0x1F;
	if (opCode == 0x00) {
	if (funct == 0x00)
	return InstructionStatistics.CATEGORY_ALU; // sll
	if (0x02 <= funct && funct <= 0x07)
	return InstructionStatistics.CATEGORY_ALU; // srl, sra, sllv, srlv, srav
	if (funct == 0x08 || funct == 0x09)
	return InstructionStatistics.CATEGORY_JUMP; // jr, jalr
	if (0x10 <= funct && funct <= 0x2F)
	return InstructionStatistics.CATEGORY_ALU; // mfhi, mthi, mflo, mtlo, mult, multu, div, divu, add, addu, sub, subu, and, or, xor, nor, slt, sltu
	return InstructionStatistics.CATEGORY_OTHER;
	}
	if (opCode == 0x01) {
	if (0x00 <= funct && funct <= 0x07)
	return InstructionStatistics.CATEGORY_BRANCH; // bltz, bgez, bltzl, bgezl
	if (0x10 <= funct && funct <= 0x13)
	return InstructionStatistics.CATEGORY_BRANCH; // bltzal, bgezal, bltzall, bgczall
	return InstructionStatistics.CATEGORY_OTHER;
	}
	if (opCode == 0x02 || opCode == 0x03)
	return InstructionStatistics.CATEGORY_JUMP; // j, jal
	if (0x04 <= opCode && opCode <= 0x07)
	return InstructionStatistics.CATEGORY_BRANCH; // beq, bne, blez, bgtz
	if (0x08 <= opCode && opCode <= 0x0F)
	return InstructionStatistics.CATEGORY_ALU; // addi, addiu, slti, sltiu, andi, ori, xori, lui
	if (0x14 <= opCode && opCode <= 0x17)
	return InstructionStatistics.CATEGORY_BRANCH; // beql, bnel, blezl, bgtzl
	if (0x20 <= opCode && opCode <= 0x26)
	return InstructionStatistics.CATEGORY_MEM; // lb, lh, lwl, lw, lbu, lhu, lwr
	if (0x28 <= opCode && opCode <= 0x2E)
	return InstructionStatistics.CATEGORY_MEM; // sb, sh, swl, sw, swr
	return InstructionStatistics.CATEGORY_OTHER;
	}

If you want to try to solve either of these and have trouble, feel free to ask for advice.

[esperz2019]

Hello,
I think the Instruction Statistics tool is not working properly.
The amount of memory instructions is always 0 or 1, while sometimes the Others will make up about 50% of the total.
I haven't looked into it yet and would like to ask if it is a bug?

[TheThirdOne]

@esperz2019, you are correct; Instruction Statistics is not working properly. The tool was never updated from using MIPS. It reads the RISC-V instructions as if they were MIPS instructions and therefore incorrectly categorizes them.

The snippet I linked above shows exactly how the classification of instructions according to MIPS works.

I haven't yet fixed it because its a nice self contained issue that a new contributor could solve.

[esperz2019]

@TheThirdOne Thanks, I'd like to have a try if I have some time later.

[giancarlopernudisegura]

I'm currently modifying the Instruction Statistics tool to properly categorize the A extension instructions as part of #75.

rars/src/rars/tools/InstructionStatistics.java

Lines 238 to 241 in 0695981

	protected int getInstructionCategory(ProgramStatement stmt) {
	int opCode = stmt.getBinaryStatement() >>> (32 - 6);
	int funct = stmt.getBinaryStatement() & 0x1F;

After referring to the RISC-V Reference Datasheet, I believe the variables opCode and funct are reversed, as funct7 is in 25-31 and opcode is in 0-6. Furthermore, since opcode is 7 bits long, the bitmask to obtain it should be 0x7F, not 0x1F.

[TheThirdOne]

@giancarlopernudisegura, The code in InstructionStatistics is almost entirely wrong for RISC-V. I would recommend using instanceof checks rather than binary manipulation as an easier way to check classes of instructions. Classes like Arithmetic already group various types of instructions.

[TheThirdOne]

This is officially fixed now. Thanks to @esperz2019 and @giancarlopernudisegura.